Customer Demand Prediction Under Seasonal Uncertainty: Prophet vs. LSTM for Coffee Shop NetSales Forecasting

Accurate sales forecasting is critical for Small and Medium Enterprises (SMEs) in the food and beverage sector, where perishable inventory and shifting demand create operational pressures. This study compares the performance of two forecasting models, Prophet and Long Short-Term Memory (LSTM) in predicting daily NetSales for a coffee shop during the Ramadhan and Ied periods. Prophet offers interpretable outputs and accommodates holiday and seasonal patterns, while LSTM, a deep learning model, captures complex, non-linear time-based relationships. Using transaction data from 2024 and early 2025, both models were assessed using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). LSTM demonstrated greater accuracy, particularly in reflecting the irregular sales patterns typical of Ramadan. These results support more precise planning for inventory and staffing, helping SMEs respond effectively to seasonal fluctuations in consumer demand.